Sean: I was just speaking with one of the Mellanox guys at the booth and he said that there is a parameter that performs a similar function to the TCP backlog that is set in the MAD message used to advertise the service point. He couldn't remember the name, but said I should post it to this group as a question.
So the question is "what is the method used to manage the number of outstanding, uncompleted connection requests coming in to a single service point? Is there some parameter you can specify in the MAD messages used to set up the connection?" WRT the CMA: I think this parameter is just an attribute of the service/listening endpoint and not a queuing depth of outstanding, unaccepted/rejected connections in the CMA. On Wed, 2005-11-16 at 11:54 -0800, Sean Hefty wrote: > I've been trying to implement support for a backlog parameter in the CMA's > rdma_listen() call. My original goal was to push the backlog parameter down > into the kernel CMA with the following definition: > > Backlog - The number of unprocessed connection requests that a listener can > have. A connection request is considered processed once it has either been > accepted or rejected. > > To keep things simple, is there any reason for the kernel CMA to have a > backlog > parameter? Would iWarp require this? The kernel CMA is callback driven, > which > makes it trivial for clients to manage their own backlog. I'm considering > maintaining this within the kernel uCMA only, unless it would need to be > pushed > down for iWarp. > > - Sean > > _______________________________________________ > openib-general mailing list > [email protected] > http://openib.org/mailman/listinfo/openib-general > > To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general _______________________________________________ openib-general mailing list [email protected] http://openib.org/mailman/listinfo/openib-general To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general
